KR102665709B1 - Artificial intelligence-based smart metering prediction management system - Google Patents

Abstract

The present invention relates to an artificial intelligence-based smart metering prediction management system. More specifically, an artificial intelligence system that can provide water usage prediction information, freeze-thaw prediction information, and indoor water leakage prediction information using an artificial intelligence-based prediction model for each analysis object. It is about a smart metering-based predictive management system. For this purpose, the artificial intelligence-based smart metering prediction management system collects water meter information for each point in the area of interest and weather information for the area of interest by period, and has a data acquisition unit that acquires sample data for each period, and detects data from the water meter information for each point. A data management unit that identifies the sample data for each period as either valid data or invalid data based on metering error missing data, and as identified as valid data, the sample data for each period is divided into past data and future data based on a specific date. A data division unit that divides the data, a data learning unit that learns an artificial intelligence-based prediction model for each analysis object through machine learning with the past data as input and the future data as output, a prediction model for each analysis object, and sample data for each period. , It includes an integrated service unit that provides GIS-based control information to a plurality of customer terminals through a web service based on previously collected facility DB information and previously collected map DB information.

Description

Artificial intelligence-based smart metering prediction management system{ARTIFICIAL INTELLIGENCE-BASED SMART METERING PREDICTION MANAGEMENT SYSTEM}

The present invention relates to an artificial intelligence-based smart metering prediction management system. More specifically, an artificial intelligence system that can provide water usage prediction information, freeze-thaw prediction information, and indoor water leakage prediction information using an artificial intelligence-based prediction model for each analysis object. It is about a smart metering-based predictive management system.

Recently, in order to maximize management efficiency and asset management effectiveness of water supply facilities, demand for technologies such as remote measurement using drones/unmanned aerial vehicles, 3D data construction, GIS combination, and AI analysis is increasing.

However, existing water supply systems are limited to one-dimensional demand forecasting and supply control using data such as low-density sensor networks and rate systems, and are managed based on inaccurate forecasts.

In particular, the integrated management system for water supply facility management is based on two-dimensional planar data, and is pointed out as a cause of reducing the usability and reliability of the existing GIS system due to missing and inaccurate data.

Therefore, in order to develop a water production plan, pump operation plan, and water rate system suitable for the domestic situation, smart metering prediction management linked to AI-based big data analysis such as water usage pattern analysis, freezing and bursting pattern analysis, and indoor water leakage pattern analysis. System development is needed.

The present invention is to solve the above problems, and the purpose of the present invention is to provide an artificial intelligence-based system that can provide water usage prediction information, freeze-thaw prediction information, and indoor water leakage prediction information using an artificial intelligence-based prediction model for each analysis target. It is intended to provide a smart metering predictive management system.

The above and other objects and advantages of the present invention will become apparent from the following description of preferred embodiments.

To achieve the above purpose, the artificial intelligence-based smart metering prediction management system according to an embodiment of the present invention collects water meter information for each point in the area of interest and weather information for the area of interest by period to obtain sample data for each period. A data acquisition unit that identifies the sample data for each period as either valid data or invalid data, based on metering error missing data detected from the water meter information for each point, a data management unit that identifies sample data for each period as valid data or invalid data, A data division unit that divides sample data into past data and future data based on a specific date, and data that learns a prediction model for each analysis target based on artificial intelligence through machine learning with the past data as input and the future data as output. Based on the learning unit and the prediction model for each analysis target, sample data for each period, previously collected facility DB information, and previously collected map DB information, GIS-based control information is provided to multiple customer terminals through a web service. It includes a service unit, and the GIS-based control information includes water usage prediction information, freeze and burst prediction information, and indoor water leakage prediction information for the area of interest.

According to an embodiment of the present invention, by providing water usage prediction information, freezing and bursting prediction information, and indoor water leakage prediction information using an artificial intelligence-based prediction model for each analysis object, it supports more efficient management of water supply business operations. It can prevent and reduce social costs caused by freezing and indoor water leaks.

Figure 1 is a diagram schematically showing an artificial intelligence-based smart metering prediction management system 1000 according to an embodiment of the present invention.
Figure 2 is an example diagram of sample data for each period.
Figures 3(A) and 3(B) are examples to explain missing metering data and high-quality data through AI quality control.
FIG. 4A is a block diagram specifically showing the data management unit 200 of FIG. 1, FIG. 4B is an example of a sample chart graph of an area of interest, FIG. 4C is an example of pollution information, and FIG. 4D is a data management unit by time and day of the week. These are examples of water use pattern distribution charts.
FIG. 5 is a block diagram showing one embodiment of the data learning unit 400 of FIG. 1.
Figures 6(A) to (C) show examples for selecting a time series prediction model in advance.
FIG. 7 is a block diagram showing another embodiment of the data learning unit 400_1 of FIG. 1.
8(A) to 8(C) show embodiments for selecting a loss value correction model in advance.
FIG. 9 is a block diagram showing an embodiment of the integrated service unit 500 of FIG. 1.

Hereinafter, the present invention will be described in detail with reference to embodiments of the present invention and drawings. These examples are merely presented as examples to explain the present invention in more detail, and it will be apparent to those skilled in the art that the scope of the present invention is not limited by these examples. .

Additionally, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by a person skilled in the art to which the present invention pertains, and in case of conflict, this specification including definitions The description will take precedence.

In order to clearly explain the proposed invention in the drawings, parts unrelated to the description have been omitted, and similar reference numerals have been assigned to similar parts throughout the specification. And, when it is said that a part "includes" a certain component, this means that it does not exclude other components, but may further include other components, unless specifically stated to the contrary. Additionally, “unit” as used in the specification refers to a unit or block that performs a specific function.

Identification codes (first, second, etc.) for each step are used for convenience of explanation. The identification codes do not describe the order of each step, and each step does not clearly state a specific order in context. It may be carried out differently from the order specified above. That is, each step may be performed in the same order as specified, may be performed substantially simultaneously, or may be performed in the opposite order.

Figure 1 is a diagram schematically showing an artificial intelligence-based smart metering prediction management system 1000 according to an embodiment of the present invention, Figure 2 is an exemplary diagram of sample data by period, and Figures 3(A) and 3(B) ) is an example diagram to explain metering error-missing data and high-quality data through AI quality control.

1 to 3(B), the artificial intelligence-based smart metering prediction management system 1000 includes a data acquisition unit 100, a data management unit 200, a data division unit 300, and a data learning unit. (400) and may include the integrated service department (500).

First, the data acquisition unit 100 can acquire sample data for each period by collecting water meter information for each point in the area of interest and weather information for the area of interest by period.

Here, sample data for each period may include water meter value, air temperature, humidity, ground temperature, and freeze rupture potential index, as shown in FIG. 2.

In other words, the water meter information for each point in the area of interest is a point-by-point list of water meter values, and the weather information for the area of interest includes the temperature, humidity, ground temperature, and freeze potential index of the area of interest transmitted from the Korea Meteorological Administration server 12. You can.

For example, the data acquisition unit 100 receives water meter information for each point and the Korea Meteorological Administration server 12 received from a plurality of IoT sensors (11_1 to 11_N) installed for each point in order to measure the water meter value for each point in the area of interest. The weather information received from can be classified into sample data for each period and stored in the storage DB (700).

The data acquisition unit 100 may collect in advance facility DB information for each point in the area of interest and map DB information for each point in the area of interest and store them in the storage DB 700.

According to the embodiment, the data acquisition unit 100 determines the meter reading status of the water meter information for each point in the area of interest based on whether there is a correspondence between the water meter information for each point in the area of interest and the preset standard information for each meter reading state. You can tag water meter information for each location.

For example, based on whether there is a correspondence between the water meter information for each point in the area of interest and the preset freezing and rupture standard information, the data acquisition unit 100 determines the freeze and rupture status for the water meter information for each point in the area of interest and calculates the freeze and rupture status for each point. You can tag water meter information.

Next, the data management unit 200 can identify and classify sample data for each period as either valid data or invalid data, based on metering error missing data detected from water meter information for each point in the area of interest.

Here, as shown in FIG. 3(A), metering incorrect missing data may mean water meter information with a change value greater than a preset value over a certain period of time.

At this time, high-quality data through AI quality control may mean water meter information with change values below a preset value for a certain period of time, as shown in FIG. 3(B).

Next, as the sample data for each period is identified as valid data, the data division unit 300 may divide the sample data for each period into past data and future data based on a specific date.

According to one embodiment, the data division unit 300 may remove a period identified as invalid data from sample data for each period and then identify the remaining sample data as valid data.

According to another embodiment, the data division unit 300 may set a specific date as a reference date that corresponds to the weather information of the area of interest and corresponds to a previous date within a certain period of time from the present.

Next, the data learning unit 400 can learn an artificial intelligence-based prediction model for each analysis target through machine learning using past data as input and future data as output.

Here, the artificial intelligence-based prediction model for each analysis object may include a water usage prediction model, a freeze-and-break prediction model, and an indoor water leakage prediction model.

For example, the prediction model for each analysis object may be one of DeePAR and GRU based on a recurrent neural network, or Transformer and TFT based on a self-attention mechanism. Specifically, DeePAR and GRU can derive more accurate output values, which is an advantage of recurrent neural networks, and Transformer and TFT can be deep learning algorithms that process large-scale data more efficiently, which is an advantage of self-attention.

Next, the integrated service unit 500 connects a plurality of customer terminals (10_1 to 10_N) through a web service based on the prediction model for each analysis object, sample data for each period, previously collected facility DB information, and previously collected map DB information. GIS-based control information can be provided.

Here, the GIS-based control information may include water usage prediction information, freeze and burst prediction information, and indoor water leakage prediction information for the area of interest. Additionally, a web service may be a web page provided through a web server.

According to one embodiment, the integrated service unit 500 is an insulation object detected from a water supply facility image uploaded through a web service from any one customer terminal (eg, 10_1) among a plurality of customer terminals (10_1 to 10_N). Based on this, it is possible to diagnose whether the insulation object is installed normally.

For example, images of water supply facilities may be images of water pipes, water pipe valves, and water meters.

According to another embodiment, the integrated service department 500 provides water for the prediction model for each analysis object based on the scale of water supply facility failure determined according to natural disasters such as floods, droughts, typhoons, and earthquakes or events such as war. You can set usage weight.

For example, if the scale of the water supply facility failure is greater than a preset scale, the integrated service department 500 can adjust the water usage weight for the prediction model for each analysis object to reduce it. In addition, if the scale of the water supply facility failure is less than the preset scale, the integrated service department 500 can adjust the water usage weight for the prediction model for each analysis object to increase the weight.

According to another embodiment, the integrated service unit 500 determines the freezing and damage prediction weight for the prediction model for each analysis object based on the deterioration condition analyzed through image analysis of a plurality of facility images for water supply facilities in the area of interest. can be set.

For example, when the deterioration state is higher than a preset normal state, the integrated service unit 500 can adjust the freeze-thaw prediction weight for the prediction model for each analysis object to increase it. In addition, when the deterioration state is less than the preset normal state, the integrated service unit 500 can adjust the freeze prediction weight for the prediction model for each analysis object in a direction to reduce it.

According to another embodiment, the integrated service unit 500 predicts freezing and damage based on location information received through a web service from any one customer terminal (e.g., 10_1) among a plurality of customer terminals 10_1 to 10_N. A route guidance inspection map can be provided to inspect the frozen and burst predicted points extracted from the information and the indoor water leak predicted points extracted from the indoor water leakage prediction information.

According to another embodiment, the integrated service unit 500 provides a freeze prevention method in the form of a pop-up message when the location information transmitted through a web service from any one customer terminal (e.g., 10_1) corresponds to a freeze prediction point. The service can be provided to any one customer terminal (eg, 10_1) through a web service.

Here, the freeze prevention method may include regular inspection of water supply equipment, cleaning around the water supply equipment, opening the water supply equipment valve, and installing insulation.

According to another embodiment, the integrated service unit 500 provides a region-of-interest prediction information service video based on a plurality of images that are automatically captured and synthesized as it services water usage prediction information, freeze and burst prediction information, and indoor water leakage prediction information. can be provided for download through a web service.

Hereinafter, the configuration of the present invention and its effects will be described in more detail through specific examples and comparative examples. However, these examples are for illustrating the present invention in more detail, and the scope of the present invention is not limited to these examples.

FIG. 4A is a block diagram specifically showing the data management unit 200 of FIG. 1, FIG. 4B is an example of a region of interest sample chart graph, FIG. 4C is an example of incorrect missing information, and FIG. 4D is a time and These are examples of water use pattern distribution by day of the week.

1 and 4A to 4D, the data management unit 200 may include a drawing unit 210, a period filtering unit 220, and a usage analysis unit 230.

First, the drawing unit 210 can draw a sample chart graph of an area of interest using sample data for each period.

Here, the area of interest sample chart graph may be a graph showing water meter values, air temperature, humidity, ground temperature, and freeze rupture potential index over a certain period of time, as shown in FIG. 4B.

Next, the period filtering unit 220 may remove the period in which erroneous missing occurs from the sample data for each period based on the erroneous missing information confirmed from the region of interest sample chart graph.

Here, as shown in FIG. 4C, the incorrect missing information may mean sample data having a bounce value with a preset numerical change, an initial state change value, a period error value, and a no change value.

Next, when the sample data for each period is identified as valid data, the usage analysis unit 230 selectively generates a water use pattern distribution chart by time and day through the sample data for each period from which the erroneous occurrence period has been removed, and a plurality of customer terminals ( It can be provided to 10_1~10_N).

Here, the water use pattern distribution chart by time and day of the week is data for confirming the usage pattern, as shown in FIG. 4D, and can be used as learning material for a prediction model for each analysis object or provided directly to customers.

Figure 5 is a block diagram showing an embodiment of the data learning unit 400 of Figure 1, and Figures 6(A) to (C) show embodiments for selecting a time series prediction model in advance.

When described with reference to FIGS. 1, 5, and 6, the data learning unit 400 may include a probability distribution estimation unit 410 and an error filtering unit 420.

First, the probability distribution estimation unit 410 can estimate the probability distribution of the first predicted value by individually learning valid data through a time series prediction model. Additionally, the probability distribution estimation unit 410 can estimate the probability distribution of the second predicted value by individually learning the error data through a time series prediction model.

Here, error data refers to data in which a specific numerical value from valid data is transformed beyond a preset value,

At this time, the first predicted value probability distribution is a function that represents the probability of having a specific value as valid data is applied to the time series prediction model, and the second predicted value probability distribution is a function that represents the probability of having a specific value as error data is applied to the time series prediction model. It may be a function that represents

Additionally, the time series prediction model can be either DeePAR or GRU based on recurrent neural network.

Next, the error filtering unit 420 may pre-filter the abnormal data from the valid data based on the predicted value confidence interval for the valid data determined according to the difference between the first and second predicted value probability distributions. That is, the error filtering unit 420 can output valid data from which abnormal data has been removed.

Depending on the embodiment, the probability distribution estimation unit 410 selects at least one time series prediction model in advance based on the loss function output change value for each model analyzed through a preset loss function for a plurality of time series prediction model candidates. can do.

Here, the preset loss function may use Gaussian Negative Log Likelihood (GNLL), as shown in FIG. 6(A).

At this time, the loss function output change value for each model can be analyzed according to time series prediction model candidates and displayed in a graph, as shown in FIG. 6(B).

In addition, the plurality of time series prediction model candidates may include DeePAR and GRU based on recurrent neural network, and Transformer and TFT based on self-attention mechanism, as shown in FIG. 6(C).

According to another embodiment, the probability distribution estimation unit 410 may manually switch the time series prediction model based on the data capacity of the valid data.

For example, if the valid data exceeds the preset data capacity, the probability distribution estimator 410 may select DeePAR as the time series prediction model. Additionally, if the valid data does not exceed the preset data capacity, the probability distribution estimation unit 410 may select the time series prediction model as GRU.

FIG. 7 is a block diagram showing another embodiment of the data learning unit 400_1 of FIG. 1, and FIGS. 8(A) to 8(C) show embodiments for selecting a loss value correction model in advance.

1, 5, 7, and 8, the data learning unit 400_1 includes a probability distribution estimation unit 410, an error filtering unit 420, a defect learning unit 430, and a correction processing unit 440. ) may include.

Hereinafter, duplicate descriptions of the probability distribution estimation unit 410 and the error filtering unit 420 of the same member number described in FIG. 5 will be omitted.

First, the defect learning unit 430 can learn a loss correction model through machine learning using previously collected defect data as input and the probability distribution of the first predicted value for valid data as output.

Here, defective data may mean data in which a portion of data from valid data has been lost.

At this time, the loss value correction model may be either SAITS or Transformer based on a self-attention mechanism.

Next, if some sections of valid data are missing, the correction processing unit 440 may preprocess the valid data into correction data through a loss value correction model.

Depending on the embodiment, the defect learning unit 430 may determine at least one loss correction model based on the loss function output change value for each model analyzed through a preset loss function for a plurality of loss correction model candidates. there is.

Here, the preset loss function can use the Mean Squared Error (MSE), which is the sum of the squares of the difference between the actual value and the predicted value, as shown in FIG. 8(A).

At this time, the loss function output change value for each model can be displayed in a graph as the loss function output value calculated by individually learning the loss correction model candidates, as shown in FIG. 8(B).

In addition, the plurality of loss correction model candidates may include SAITS and Transformer based on a self-attention mechanism, and BRITS and MRNN based on a recurrent neural network, as shown in FIG. 8(C).

FIG. 9 is a block diagram showing an embodiment of the integrated service unit 500 of FIG. 1.

1 and 9, the integrated service unit 500 includes an input unit 510, a data extraction unit 520, a model application unit 530, a prediction information generation unit 540, and a page service unit 550. ) may include.

First, the input unit 510 may receive a local keyword input from any one of the plurality of customer terminals 10_1 to 10_N (eg, 10_1) through a web-based meter reading prediction service.

Here, the web-based meter reading prediction service may be an application program provided and installed on a plurality of customer terminals (10_1 to 10_N) connected through a web server.

Next, the data extraction unit 520 can extract regional valid data corresponding to regional keywords from sample data for each period.

Depending on the embodiment, the data extraction unit 520 provides a customer terminal (e.g., 10_1) through a meter reading prediction service so that branch areas listed in the regional keyword can be selected based on the data capacity information of the regional valid data. You can request it.

Next, the model application unit 530 applies regional valid data to the prediction model for each analysis object learned through the data learning unit 400, so that water usage, freezing, bursting, and indoor water leakage prediction data can be individually derived.

Next, the prediction information generation unit 540 may analyze water usage, freezing and bursting and indoor water leakage prediction data and regional effective data to generate water usage service information, freezing and breaking service information and indoor water leakage service information.

Specifically, water usage service information may include water usage history information, water usage forecast information, meta information related to regional effective data, and real-time water usage information tagged on a map. In addition, the freeze rupture service information may include real-time temperature information, a freeze rupture prediction list, and a freeze rupture risk prediction graph tagged on a map divided by freeze rupture risk stage. In addition, the indoor water leak service information may include the number of real-time detections of indoor water leaks displayed on the map, a list of indoor water leak predictions, and a comparison graph between water usage and estimated indoor water leakage.

Next, the page service unit 550 creates individual web pages corresponding to water usage service information, freezing and bursting service information, and indoor water leakage service information and provides them to one customer terminal (e.g., 10_1) through a meter reading prediction service. can do.

In this specification, only a few examples of various embodiments implemented by the present inventors are described, but the technical idea of the present invention is not limited or limited thereto, and of course, it can be modified and implemented in various ways by those skilled in the art.

10_1~10_N: Multiple customer terminals
100: Data acquisition department
200: Data Management Department
300: Data division unit
400: Data Learning Department
500: Integrated Services Department
1000: Artificial intelligence-based smart metering predictive management system

Claims (8)

A data acquisition unit that collects water meter information by point in the area of interest and weather information in the area of interest by period to obtain sample data by period;
a data management unit that identifies the sample data for each period as either valid data or invalid data based on metering error missing data detected from the water meter information for each point;
a data dividing unit that divides the sample data for each period into past data and future data based on a specific date as it is identified as valid data;
A data learning unit that learns an artificial intelligence-based prediction model for each analysis target through machine learning using the past data as input and the future data as output; and
It includes an integrated service unit that provides GIS-based control information to a plurality of customer terminals through a web service based on the prediction model for each analysis object, sample data for each period, previously collected facility DB information, and previously collected map DB information. do,
The GIS-based control information includes water usage prediction information, freeze and burst prediction information, and indoor water leakage prediction information for the area of interest,
The integrated service unit is an artificial intelligence unit that diagnoses whether the insulation object is properly installed based on the insulation object detected from the water supply facility image uploaded through the web service from any one of the plurality of customer terminals. Intelligence-based smart metering predictive management system. delete According to paragraph 1,
The data management unit includes a graph drawing unit that draws a sample chart graph of an area of interest using the sample data for each period;
a period filtering unit that removes a period in which erroneous missing occurs from the sample data for each period based on the erroneous missing information identified from the region of interest sample chart graph; and
When the sample data for each period is identified as valid data, it includes a pattern derivation unit for deriving a water use pattern distribution by time and day through the sample data for each period from which the erroneous occurrence period has been removed,
The incorrect missing information refers to a period with a bounce value, initial state change value, period error value, and no change value with a preset numerical change. An artificial intelligence-based smart metering prediction management system. According to paragraph 1,
The data learning unit includes a probability distribution estimation unit that individually estimates first and second predicted value probability distributions by individually learning the valid data and error data through a time series prediction model; and
An error filtering unit that pre-filters abnormal data from the valid data based on a confidence interval of the predicted value for the valid data determined according to the difference between the first and second predicted value probability distributions,
Here, error data refers to data in which a specific numerical value from valid data is transformed beyond a preset value,
The first predicted value probability distribution is a function representing the probability of having a specific value as the valid data is applied to the time series prediction model, and the second predicted value probability distribution is a function that represents the probability of having a specific value as the error data is applied to the time series prediction model. An artificial intelligence-based smart metering prediction management system, which is a function that represents the probability of having . According to clause 4,
The data learning unit includes a defect learning unit that learns a loss value correction model through machine learning using previously collected defect data as input and a probability distribution of a first predicted value for the valid data as output; and
When a partial section of the valid data is missing, it further includes a correction processing unit that preprocesses the valid data into correction data through the loss value correction model,
The defective data refers to data in which some sections of data are missing from the valid data. An artificial intelligence-based smart metering prediction management system. According to clause 4,
The probability distribution estimator preselects the time series prediction model based on the loss function output change value for each model analyzed through a preset loss function for a plurality of time series prediction model candidates,
The preset loss function uses Gaussian Negative Log Likelihood (GNLL),
The multiple time series prediction model candidates include DeePAR and GRU based on recurrent neural network, Transformer and TFT based on self-attention mechanism, and an artificial intelligence-based smart metering prediction management system. According to clause 5,
The defect learning unit determines the loss correction model in advance based on the loss function output change value for each model analyzed through a preset loss function for a plurality of loss correction model candidates,
The preset loss function uses Mean Squared Error (MSE), which is the sum of the squares of the difference between the actual value and the predicted value.
The plurality of loss value correction model candidates include SAITS and Transformer based on a self-attention mechanism, and BRITS and MRNN based on recurrent neural network, an artificial intelligence-based smart metering prediction management system. According to paragraph 1,
The integrated service unit includes an input unit that receives a local keyword from any one of a plurality of customer terminals through the web service;
a data extraction unit that extracts regional valid data corresponding to regional keywords from the sample data for each period;
A model application unit that individually derives water usage, freezing, bursting, and indoor water leakage prediction data by applying the regional effective data to the prediction model for each analysis object;
a prediction information generator that analyzes the water usage, freezing and bursting and indoor water leakage prediction data and the regional effective data to generate water usage service information, freezing and breaking service information and indoor water leakage service information; and
An artificial intelligence-based smart metering prediction that includes a page service unit that generates individual web pages corresponding to each of the water usage service information, the freezing service information, and the indoor water leakage service information and serves them to any one customer terminal so that they can be downloaded. Management system.